Soroush Vosoughi

I work on NLP and machine learning, with a focus on understanding, evaluating, and improving large language models. My interests include LLM interpretability, behavioral evaluation, alignment, bias and toxicity, misinformation and persuasion, multimodal reasoning, grounding, metacognition, computational social science, neurosymbolic AI, and cognitively feasible models of reasoning. I am particularly interested in methods and benchmarks that expose model failure modes, explain internal behavior, and support more transparent, reliable, and human-compatible AI systems.